Skip to content

Creating Sessions

Sessions are where the magic happens. This guide walks you through creating a new session with the right configuration for your task.

  1. Click New Session from the Dashboard or Sessions page
  2. Follow the session creation wizard

Choose a prompt template that matches your task:

  • Code Review - Analyze code for issues and improvements
  • Bug Fix - Investigate and fix reported bugs
  • Feature Implementation - Build new features from requirements
  • Refactoring - Improve code structure
  • Custom - Start with a blank slate

Use the search and tag filters to find the right workflow. Tags like frontend, backend, devops help narrow down options.

Select which AI model powers your session:

ProviderModelsBest For
AnthropicClaude 3.5 Sonnet, Claude 3 OpusComplex reasoning, code generation
OpenAIGPT-4, GPT-4 TurboGeneral tasks, broad knowledge

If your task involves a specific codebase:

  1. Click Connect Repository
  2. Select from your GitHub repositories
  3. Choose the branch to work from

The repository will be cloned into your session’s workspace, giving the agent full access to the code.

Expand Advanced Options for additional configuration:

  • Session Name - Custom name for easy identification
  • Initial Message - Pre-fill the first message to the agent
  • Environment Variables - Set custom env vars for the session

Click Start Session to launch. Cheffed will:

  1. Provision an isolated container
  2. Clone your repository (if connected)
  3. Initialize the agent with your chosen prompt
  4. Open the chat interface

This typically takes 10-30 seconds.

Each session runs in an isolated container with:

  • Full shell access - Run any CLI commands
  • File system - Read, write, and edit files
  • Git - Full git capabilities
  • Package managers - npm, pip, and other tools pre-installed
  • Your repository - Cloned and ready to work

Sessions remain active while you’re interacting with them. After 10 minutes of inactivity:

  1. The container spins down to save resources
  2. Session state and history are preserved
  3. Reopening the session spins up a fresh container
  4. Your conversation history remains intact
  1. Be specific - Clear, detailed prompts get better results
  2. Provide context - Share relevant files, error messages, or requirements
  3. Iterate - Use follow-up messages to refine the agent’s work
  4. Review changes - Always review code changes before applying them